writing
Skill1.1k starsChanged 5 days agoApache-2.0View on GitHub
Write prose people will actually read. Use for any prose you produce - docs, READMEs, release notes, blog posts, PR descriptions, issue text, commit bodies, design docs, emails, chat answers. Applies Orwell's six rules and strips the LLM tells - passive voice, dead metaphors, and -ly padding.
55 other projects use a version of this, 2 of them edited.
Trending
Less talk lately.about the whole repository
Most recently taken up by
- mblauberg/provenanttheir own version, first seen Nov 16, 2026
- Protonicwave/claude-skill-writingtheir own version, first seen Nov 16, 2026
- wxxb789/hoatheir own version, first seen Nov 16, 2026
- ryanportfolio/Harness-Firmwaretheir own version, first seen Nov 15, 2026
- jameswyse/skillstheir own version, first seen Nov 15, 2026
- iliaal/ai-skillstheir own version, first seen Nov 14, 2026
Who's talking about agentic-community/mcp-gateway-registry
- ards-project/ard-spec“Open-source reference implementation: MCP Gateway Registry (Apache-2.0) — both Catalog Publisher and Registry”GitHub · Jun 23, 2026 · 10 reactions
- aaif/project-proposals“[Project Proposal] MCP Gateway Registry”GitHub · May 12, 2026 · 4 reactions
- openai/codex“MCP Server Registry Discovery: connect Codex to an enterprise MCP server catalog with a single command”GitHub · Jun 2, 2026 · 3 reactions
- A Hacker News story“MCP Gateway and Registry: Enterprise-Grade Tool Governance for AI Agents”Hacker News · Oct 29, 2025 · 2 points
- anthropics/claude-code“[FEATURE] MCP Server Registry Discovery: connect Claude Code to an enterprise MCP server catalog”GitHub · Jun 2, 2026 · 1 reactions
The biggest 5 of 15 mentions linking to the repository that our crawl found. Each line is their own headline.
See all 15
Talked about on GitHub, Hacker News and Bluesky (15 times).
On GitHub · 13
- Jun 23, 2026Open-source reference implementation: MCP Gateway Registry (Apache-2.0) — both Catalog Publisher and Registry — ards-project/ard-spec10 reactions · 11 comments
- May 12, 2026[Project Proposal] MCP Gateway Registry — aaif/project-proposals4 reactions · 8 comments
- Jun 2, 2026MCP Server Registry Discovery: connect Codex to an enterprise MCP server catalog with a single command — openai/codex3 reactions · 1 comment
- Jun 2, 2026[FEATURE] MCP Server Registry Discovery: connect Claude Code to an enterprise MCP server catalog — anthropics/claude-code1 reaction · 8 comments
- Oct 24, 2025Add --header CLI argument to support arbitrary HTTP headers for MCP Gateway authentication — cisco-ai-defense/mcp-scanner1 reaction · 7 comments
8 more on GitHub
- Oct 26, 2026Scaffold 12 onboarding target(s) (disabled — vet + enable) — azmartone67/dchub-mcp-server0 reactions
- Oct 21, 2026[GAP-0056] Check MCP tool names, descriptions and schemas for injected instructions, and hide only the tools that fail — DICOMPUTE/Open0 reactions · 2 comments
- Sep 16, 2026P2 CANDIDATE / NOT EXECUTION-AUTHORIZED — Evaluate MCP Gateway Registry as unified inference+MCP+A2A router — pashamango/minutta-rp-pr-cycle-lab0 reactions
- Aug 25, 2026[New Service] August 26 OSS catalog additions — haoruilee/awesome-agent-native-services0 reactions
- Jun 23, 2026[Dream Cycle 2026-06-23] memory: semantic drift from repeated summarization cycles — AgentDB has no governance layer + plugins,automation scan — ruvnet/ruflo0 reactions
- Jun 12, 2026Competitive positioning: Pontifex is library-embedded governance, not an MCP gateway — chris-dare/pontifex0 reactions
- Nov 12, 2025Fix ECS deployment issue — gauravrele87/mcp-gateway-registry0 reactions · 1 comment
- Jun 24, 2025[Spike] Investigate A2A/MCP Gateway agentgateway.dev — caipe-io/ai-platform-engineering0 reactions · 5 comments
On Hacker News · 1
- Oct 29, 2025MCP Gateway and Registry: Enterprise-Grade Tool Governance for AI Agents2 points · 1 comment
On Bluesky · 1
- Jul 16, 2025これは興味あるな… ほしいもんなこういう系統のなにか 📑 https://github.com/agentic-community/mcp-gateway-registry 🏷️MCP — @vvakame.bsky.social1 likes and repost
Used with
Projects with it also use addskill, apply-firmware and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
ryanportfolioSkilladdskill
Create, import, update, or install repo or personal skills: runtime ownership, resources, discovery checks.
Also in 6 projects that use this
ryanportfolioSkillapply-firmware
Bring current folder (plain folder, existing repo, old or partial firmware copy) up to full Harness Firmware in place. Mirror external repo → adopt-repo.
Also in 6 projects that use this
- +10 more from ryanportfolio
Other versions55 projects use a version of this.
Word for word · 0
None found.
Edited · 2
- aws-samples/sample-agentic-coding-harness-benchmarks★ 2094% same
- dangreenberggit/skills★ 022% same
Wrote their own version · 53
- wu-yc/LabClaw★ 1.1ksame name
- solana-foundation/explorer★ 856same name
- notque/vexjoy-agent★ 450same name
- notque/claude-code-toolkit★ 448same name
- zeuikli/claude-code-workspace★ 155same name
- joshuadavidthomas/agent-skills★ 49same name
- iliaal/ai-skills★ 45same name
- alxyrgin/secondbrain_ai★ 17same name
- echris6/ai-business-skills★ 16same name
- JHostalek/dotclaude★ 11same name
43 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- docs/llms.txtllms.txt
- agentcore-registerSkill
- benchmark-reportSkill
- create-milestoneSkill
- debugSkill
- explainerSkill
- generate-agent-cardSkill
- generate-server-cardSkill
- infra-syncSkill
- macos-setupSkill
- new-feature-designSkill
- pr-reviewSkill
- release-notesSkill
- search-benchmarkSkill
- search-registrySkill
- terraform-setupSkill
- usage-reportSkill
- usage-report-v0Skill
The file
Show its rules and text
We split this skill into the rules it gives an agent
52 rules. 0 are common, 47 are distinctive, and 5 are very specific to this project.
0
Common
Most of the 49 edited versions we read keep these.
47
Distinctive
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 12
If you want people to read your stuff, follow Orwell's rules.
- Distinctive: line 18
Never use a metaphor, simile, or other figure of speech which you are used to seeing in print.
- Distinctive: line 19
Never use a long word where a short one will do.
- Distinctive: line 20
If it is possible to cut a word out, always cut it out.
- Distinctive: line 21
Never use the passive where you can use the active.
- Distinctive: line 22
Never use a foreign phrase, a scientific word, or a jargon word if you can think of an everyday English equivalent.
- Distinctive: line 35
Use a plain statement, or invent an image that fits the thing you are describing.
- Distinctive: line 48
Delete throat-clearing openers.
- Distinctive: line 64
Keep the passive only when the actor is unknown or truly beside the point: "The row was deleted at some point before the migration."
- Very specific: line 70
Keep the technical term when it is the precise name of the thing.
- Distinctive: line 72
Write out an acronym on first use, then use it.
- Distinctive: line 76
Read the sentence aloud. If no one would say it, rewrite it.
- Distinctive: line 82
No em-dashes at all, anywhere, and no double hyphen standing in for one.
- Distinctive: line 83
No lists of three when two facts will do.
- Distinctive: line 84
No summary paragraph that repeats what you just said.
- Distinctive: line 85
No praise of the reader, the code, or yourself.
- Distinctive: line 89
Do not hedge twice. "may possibly" -> "may".
- Distinctive: line 90
Use present tense for how the system behaves, past tense for what you did.
- Distinctive: line 91
Never open with "In today's fast-paced world" or any variant.
- Distinctive: line 97
No antithesis. Do not pair "X, but Y" or "not X, rather Y" for rhythm.
- Distinctive: line 100
No corrective negation. Do not define a thing by first saying what it is not.
- Distinctive: line 103
No contrasting pairs or negative parallelism.
- Distinctive: line 106
No negative anaphora. Do not open three sentences in a row with "No..." or "Never..." for effect.
- Distinctive: line 107
No setup/payoff or landing sentences.
- Distinctive: line 110
No parataxis for drama. Do not stack short clauses to build rhythm ("It compiles.
- Distinctive: line 111
No parallel sentence structures within a paragraph.
- Distinctive: line 112
No paragraph pinning. Do not top and tail a paragraph with the same idea to frame it.
- Distinctive: line 113
No summary beats. Do not restate the point you just made in different words.
- Distinctive: line 114
No stacked noun phrases. Break "a cloud-native observability data ingestion pipeline" into words that do work.
- Distinctive: line 115
No nominalization. Turn the noun back into its verb: "perform a validation of" -> "validate", "make a decision" -> "decide", "provide …
- Distinctive: line 119
No stage directions. Do not open a sentence by pointing at the next one: "Now look at", "Watch what happens", "Here is the thing", …
- Distinctive: line 122
No closing flourishes. "That is the whole interface", "and the rest is variations on it", "end of story", "in conclusion", "ultimately" are …
- Distinctive: line 123
No teaser openers. Do not open a section by announcing how many things are coming, and do not assert a dramatic dependency between them.
- Distinctive: line 127
No appended reassurance clause.
- Distinctive: line 132
No decorative bolding. Bold a word only to mark a real term or a genuine warning.
- Distinctive: line 133
No connective padding. Cut "that said", "with that in mind", "having said that", "it is also worth adding".
- Distinctive: line 141
No slogan pairs. Two short sentences that mirror each other and end on a bare "do not" read like a tagline.
- Distinctive: line 144
No objects acting like people.
- Distinctive: line 147
No code vocabulary in prose.
- Distinctive: line 150
No institutional register.
- Distinctive: line 154
Do not let a shorter sentence change the fact.
- Distinctive: line 161
Write full sentences, not headlines.
- Distinctive: line 168
Do not pad a reply to sound warm.
- Distinctive: line 172
Run this on every draft before you ship it.
- Very specific: line 175
Search for is/are/was/were + participle.
- Very specific: line 176
Search for ly and delete or replace each hit.
- Distinctive: line 177
Delete every phrase from the ban lists above.
- Distinctive: line 179
Run the mechanical scan below and fix every hit.
- Distinctive: line 181
Check every bullet has a subject and a verb.
- Distinctive: line 183
Check every claim against something real - a file, a command output, a number.
- Very specific: line 198
Every hit is a candidate, not a verdict.
- Very specific: line 200
When you add a new tell to this file, add its pattern to TELLS in that script so the next draft gets caught instead of reviewed.
“Common” is measured: kept by at least 6 in 10 of the 49 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Writing Skill
If you want people to read your stuff, follow Orwell's rules.
LLMs are not your friends. They write in passive voice, mix figures of speech that do not fit together, and pad every sentence with -ly words. Cut all of it.
Orwell's rules
- Never use a metaphor, simile, or other figure of speech which you are used to seeing in print.
- Never use a long word where a short one will do.
- If it is possible to cut a word out, always cut it out.
- Never use the passive where you can use the active.
- Never use a foreign phrase, a scientific word, or a jargon word if you can think of an everyday English equivalent.
- Break any of these rules sooner than say anything outright barbarous.
Rule 6 outranks the other five. A stiff sentence that obeys rules 1-5 is worse than a plain one that breaks one of them.
How to apply each rule
1. No stock figures of speech
Kill any phrase you have read a hundred times. If two images sit in one sentence, they will clash and the reader sees nothing.
Ban list, not exhaustive: game changer, at the end of the day, low-hanging fruit, moving the needle, paradigm shift, deep dive, unlock value, seamless, robust, journey, landscape, ecosystem (unless you mean living things), leverage as a verb, delve, tapestry, testament to, navigate the complexities.
Use a plain statement, or invent an image that fits the thing you are describing.
- Bad: This release is a game changer that unlocks seamless value across the ecosystem.
- Good: This release cuts registry startup from 40 seconds to 3.
2. Short words
Prefer the short word every time: use not utilize, help not facilitate, start not commence, about not approximately, use not leverage, get not obtain, show not demonstrate, need not necessitate, before not prior to, after not subsequent to, so not accordingly, most not the majority of, can not possesses the capability to.
3. Cut words
Delete every word that carries no weight. Common padding: in order to, the fact that, it should be noted that, it is important to note, as previously mentioned, in terms of, with respect to, a number of, at this point in time, basically, essentially, actually, really, very, quite, simply, just.
Delete throat-clearing openers. Start with the fact.
- Bad: It is important to note that, in terms of performance, the cache basically helps quite a lot.
- Good: The cache cuts p99 latency by half.
Aim to cut a first draft by a third. Then read it again and cut more.
4. Active voice
Name the actor, then the verb. Passive voice hides who did what, and hidden actors are how bugs and bad decisions escape review.
- Bad: The token is validated by the auth service and errors are logged.
- Good: The auth service validates the token and logs errors.
Hunt for is/are/was/were/been/being next to a past participle. Also hunt "there is", "there are", and "it is" - each one usually buries the real subject.
Keep the passive only when the actor is unknown or truly beside the point: "The row was deleted at some point before the migration."
5. Everyday English
Say the thing in words a competent reader knows. Drop the Latin and the jargon: use for example not e.g., that is not i.e., by itself not per se, the opposite not vice versa, so far not to date, roughly not circa.
Keep the technical term when it is the precise name of the thing. JWT, Fargate, idempotent, and race condition earn their place. synergy, holistic, and operationalize do not.
Write out an acronym on first use, then use it.
6. Sound like a person
Read the sentence aloud. If no one would say it, rewrite it. Break any rule above rather than write something ugly, stilted, or unclear.
Extra rules for LLM prose
- Cut -ly adverbs. Pick a stronger verb instead.
significantly improved->doubled.carefully validates->validates. - No "not only X but also Y". No "X isn't just Y - it's Z". No "this isn't just about X", in any form.
- No em-dashes at all, anywhere, and no double hyphen standing in for one. This repo bans the character outright, so there is no "reveal at the end of a sentence" exception to argue about. Use a comma, a colon, parentheses, or two sentences.
- No lists of three when two facts will do.
- No summary paragraph that repeats what you just said.
- No praise of the reader, the code, or yourself. No "great question", no "powerful and flexible".
- One idea per sentence. Short sentences beat long ones with semicolons.
- Concrete over abstract: exact numbers, file paths, symbol names, commands.
- Say what changed and what breaks. Skip the vision.
- Do not hedge twice. "may possibly" -> "may". Pick one level of certainty and own it.
- Use present tense for how the system behaves, past tense for what you did.
- Never open with "In today's fast-paced world" or any variant.
Sentence-shape tells
Word-level fixes are not enough. LLMs lean on a handful of sentence shapes that read as machine-made even when every word is plain. Kill these.
- No antithesis. Do not pair "X, but Y" or "not X, rather Y" for rhythm. Say the one thing you mean.
- Bad: The cache is not a workaround, it is the design.
- Good: The cache is the design.
- No corrective negation. Do not define a thing by first saying what it is not.
- Bad: This isn't about speed, it's about correctness.
- Good: This fixes a correctness bug.
- No contrasting pairs or negative parallelism. Drop the "not just X, but Y" and "less A, more B" frames.
- Bad: We didn't add a feature, we removed a footgun.
- Good: We removed the retry loop that double-charged users.
- No negative anaphora. Do not open three sentences in a row with "No..." or "Never..." for effect. (This list is a list, not prose.)
- No setup/payoff or landing sentences. Do not build a sentence whose only job is to tee up the next one, and do not end a paragraph on a short punchy line meant to resonate.
- Bad: There was one thing left to fix. The timeout.
- Good: The last fix was the 30-second timeout.
- No parataxis for drama. Do not stack short clauses to build rhythm ("It compiles. It ships. It works.").
- No parallel sentence structures within a paragraph. If two sentences share the same skeleton, rewrite one.
- No paragraph pinning. Do not top and tail a paragraph with the same idea to frame it.
- No summary beats. Do not restate the point you just made in different words.
- No stacked noun phrases. Break "a cloud-native observability data ingestion pipeline" into words that do work.
- No nominalization. Turn the noun back into its verb: "perform a validation of" -> "validate", "make a decision" -> "decide", "provide support for" -> "support".
- No significance flags. Do not tell the reader that something matters; give the reason and let them judge. Ban: matters more than it looks, worth noticing, worth knowing, the part worth knowing, this is the important bit, note that, importantly, it is not obvious, and that is not a small thing.
- Bad: The regex already handles the exact-match form, which is worth knowing because it is not obvious.
- Good: The regex already handles the exact-match form, so the new block is covered without changing it.
- No stage directions. Do not open a sentence by pointing at the next one: "Now look at", "Watch what happens", "Here is the thing", "Consider", "Notice that", "Let us walk through". Start with the fact.
- Bad: Now look at what this does to the config.
- Good: Two identical exact-match locations stop nginx from loading the config at all.
- No closing flourishes. "That is the whole interface", "and the rest is variations on it", "end of story", "in conclusion", "ultimately" are applause lines. Stop when the information stops.
- No teaser openers. Do not open a section by announcing how many things are coming, and do not assert a dramatic dependency between them. Start with the first thing and let the count emerge.
- Bad: Two things, and the second exists because the first failed.
- Good: This PR adds a skill that turns an issue into an explainer. It also adds a prose linter, because the skill's own output broke the writing rules.
- Announcing a count is fine when it is doing real work, as in "Four properties make this safe" followed by four named properties. It is not fine as a drum roll.
- No appended reassurance clause. Do not count the items and then add a clause asserting they are all fine. The clause carries no information and reads as padding.
- Bad: Four properties make this safe, and each one is checkable.
- Good: Four properties make this safe.
- Bad: Three facts have to line up for the bug to exist, and all three are in place today.
- Good: The bug needs three things to be true at once, and they are.
- No decorative bolding. Bold a word only to mark a real term or a genuine warning. Bolding the first clause of every paragraph is a layout tic, and once every paragraph is bold nothing is.
- No connective padding. Cut "that said", "with that in mind", "having said that", "it is also worth adding". If the next sentence follows, it does not need a ramp.
- No headline fragments. Write sentences with a subject and a verb, not telegraph headings. This is the most common way a bulleted answer stops sounding human: every bullet becomes a topic label plus a clipped noun phrase, and the reader has to reconstruct who does what.
- Bad: Supported today, two mechanisms.
- Good: This is supported today, and there are two ways to do it.
- Bad: Lifecycle: force new assets to draft, then promote by PATCH.
- Good: On lifecycle, you can force every new asset to land in draft, then promote it with a PATCH.
- A leading label is fine when it is a real heading. It is not fine as a substitute for the sentence's subject.
- Vary sentence length on purpose, not on a pattern. Mix short and long so the rhythm is unpredictable. Do not alternate long-short-long-short.
- No slogan pairs. Two short sentences that mirror each other and end on a bare "do not" read like a tagline. Say the whole thought in one sentence.
- Bad: Authorization stays the same. Consent and blast radius do not.
- Good: Authorization works the same way it always has. What changes is how much one connection can do, and who gets asked before it does it.
- No objects acting like people. A tool does not "hold" another tool, text does not "carry" a URL, and a notification does not "reach" anything. Make a person, a service, or the code the subject, and give it a verb someone would say.
- Bad: The model that reads that text now holds a tool that reaches every write-capable backend.
- Good: The model that reads that text can now call any write-capable backend.
- No code vocabulary in prose. Words from the codebase ("vaults the token", "keys credentials on the user", "elicitation-complete notification", "has no session to reach") belong in backticks or nowhere. In a sentence, say what happens in words a reader would use.
- Bad: The gateway vaults the GitHub token, and the elicitation-complete notification has no session to reach.
- Good: The gateway stores the GitHub token, and the notification that consent finished has nowhere to go.
- No institutional register. "Operators and users treat X as", "is not an enforcement control", and "rests on" sound like a policy memo. Write it the way you would explain it to a colleague at a desk.
- Bad: AGENTS.md says prompt guidance is not an enforcement control.
- Good: AGENTS.md is clear that prompt text enforces nothing.
- No "One option... Another... A third" runs. Put a short list of options in one sentence, or in a real list.
- Do not let a shorter sentence change the fact. Compressing "the gateway that accepted the token" into "the issuer's own gateway" made it wrong, because Keycloak issues the token. Re-check each rewritten claim against the source, not against your earlier draft.
Replying to a person
Everything above applies to a reply. These are the few things a reply needs that a document does not. Use them for email, chat answers, issue and PR comments, and review responses.
- Answer in the order they asked. Echo their own words in the heading of each answer so they can match your reply to their question.
- Write full sentences, not headlines. A reply is one side of a conversation, so the headline-fragment tell above hits hardest here. Read it aloud as if you were saying it to them.
- Name yourself as the actor. A recommendation belongs to someone: "I would keep production on its own cluster", not "production isolation is recommended".
- Answer the question before you qualify it. Lead with yes, no, or the recommendation, then the detail.
- State the limits you know about. A reply that only lists what works reads like a sales sheet and costs you the next conversation.
- Close with who does what next. Name the one thing you want back from them.
- Contractions are fine. So is a short sentence that sounds like speech. Formality is not the goal, clarity is.
Do not pad a reply to sound warm. No "great question", no restating their question back at them before answering, no summary paragraph at the end.
Revision pass
Run this on every draft before you ship it. That means every file, including the second and third documents you write in a session: having loaded this skill earlier in the conversation does not count as running the pass on the new file.
- Read it aloud, one sentence at a time. For each sentence, ask: would I say this, in these words, to a colleague sitting next to me? If not, rewrite it the way you would say it. Fix anything you stumble on. This step is the one that catches slogan pairs, objects acting like people, code vocabulary, and institutional register, and the scanner cannot do it for you.
- Search for is/are/was/were + participle. Flip each to active or justify it.
- Search for
lyand delete or replace each hit. - Delete every phrase from the ban lists above.
- Cut the first sentence of each paragraph if the second one already says it.
- Run the mechanical scan below and fix every hit.
- Scan by eye for the shape tells the scan cannot catch: parallel structure, paragraph pinning, summary beats, setup/payoff, parataxis, decorative bolding.
- Check every bullet has a subject and a verb. A bullet that opens with a topic label and a colon usually does not.
- Count words. Cut a third.
- Check every claim against something real - a file, a command output, a number.
- Apply rule 6 last: read once more and undo anything that now sounds wrong.
Run the pass. Loading this skill and then skipping the pass is how the tells survive into the draft. A clean scanner run is not a finished pass either: in one session a document passed the scanner with no hits, and the read-aloud step still turned up twenty sentences no person would say, plus one factual error.
Mechanical scan
A prose rule you have to remember is a rule you will skip under deadline. Run the scanner before you hand anything over, the same way you would run a linter:
uv run python scripts/prose-scan.py <file> [<file> ...]It covers the phrase-level tells and prints path:line for each hit. Add --strict to exit non-zero on any hit, which is what a generated document should be held to. The shape-level tells still need your eyes: parallel structure, paragraph pinning, summary beats, setup/payoff, parataxis, decorative bolding.
Every hit is a candidate, not a verdict. robust is fine inside a quotation and ultimately, is sometimes load-bearing. Read each one and decide. What you may not do is leave a hit unexamined.
When you add a new tell to this file, add its pattern to TELLS in that script so the next draft gets caught instead of reviewed.
Example
Before:
It should be noted that a significant number of performance improvements have been implemented in this release, which fundamentally transforms the observability landscape by leveraging a robust new telemetry pipeline that seamlessly facilitates the collection of metrics at scale.
After:
This release adds a telemetry pipeline. The collector batches metrics every 10 seconds and cuts registry CPU use by 30%.
62 words to 22. Named the actor. Gave the numbers.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

